Alexander Hammers

Automatic anatomical brain MRI segmentation combining label propagation and decision fusion.

Regions in three-dimensional magnetic resonance (MR) brain images can be classified using protocols for manually segmenting and labeling structures. For large cohorts, time and expertise requirements make this approach impractical. To achieve automation, an individual segmentation can be propagated to another individual using an anatomical correspondence estimate relating the atlas image to the target image. The accuracy of the resulting target labeling has been limited but can potentially be improved by combining multiple segmentations using decision fusion. We studied segmentation propagation and decision fusion on 30 normal brain MR images, which had been manually segmented into 67 structures. Correspondence estimates were established by nonrigid registration using free-form deformations. Both direct label propagation and an indirect approach were tested. Individual propagations showed an average similarity index (SI) of 0.754+/-0.016 against manual segmentations. Decision fusion using 29 input segmentations increased SI to 0.836+/-0.009. For indirect propagation of a single source via 27 intermediate images, SI was 0.779+/-0.013. We also studied the effect of the decision fusion procedure using a numerical simulation with synthetic input data. The results helped to formulate a model that predicts the quality improvement of fused brain segmentations based on the number of individual propagated segmentations combined. We demonstrate a practicable procedure that exceeds the accuracy of previous automatic methods and can compete with manual delineations.

Three-dimensional maximum probability atlas of the human brain, with particular reference to the temporal lobe

Probabilistic atlases of neuroanatomy are more representative of population anatomy than single brain atlases. They allow anatomical labeling of the results of group studies in stereotaxic space, automated anatomical labeling of individual brain imaging datasets, and the statistical assessment of normal ranges for structure volumes and extents. No such manually constructed atlas is currently available for the frequently studied group of young adults. We studied 20 normal subjects (10 women, median age 31 years) with high‐resolution magnetic resonance imaging (MRI) scanning. Images were nonuniformity corrected and reoriented along both the anterior‐posterior commissure (AC–PC) line horizontally and the midsagittal plane sagittally. Building on our previous work, we have expanded and refined existing algorithms for the subdivision of MRI datasets into anatomical structures. The resulting algorithm is presented in the Appendix . Forty‐nine structures were interactively defined as three‐dimensional volumes‐of‐interest (VOIs). The resulting 20 individual atlases were spatially transformed (normalized) into standard stereotaxic space, using SPM99 software and the MNI/ICBM 152 template. We evaluated volume data for all structures both in native space and after spatial normalization, and used the normalized superimposed atlases to create a maximum probability map in stereotaxic space, which retains quantitative information regarding inter‐subject variability. Its potential applications range from the automatic labeling of new scans to the detection of anatomical abnormalities in patients. Further data can be extracted from the atlas for the detailed analysis of individual structures. Hum. Brain Mapping 19:224–247,2003. ©2003 Wiley‐Liss,Inc.

Multi-atlas based segmentation of brain images: Atlas selection and its effect on accuracy

Quantitative research in neuroimaging often relies on anatomical segmentation of human brain MR images. Recent multi-atlas based approaches provide highly accurate structural segmentations of the brain by propagating manual delineations from multiple atlases in a database to a query subject and combining them. The atlas databases which can be used for these purposes are growing steadily. We present a framework to address the consequent problems of scale in multi-atlas segmentation. We show that selecting a custom subset of atlases for each query subject provides more accurate subcortical segmentations than those given by non-selective combination of random atlas subsets. Using a database of 275 atlases, we tested an image-based similarity criterion as well as a demographic criterion (age) in a leave-one-out cross-validation study. Using a custom ranking of the database for each subject, we combined a varying number n of atlases from the top of the ranked list. The resulting segmentations were compared with manual reference segmentations using Dice overlap. Image-based selection provided better segmentations than random subsets (mean Dice overlap 0.854 vs. 0.811 for the estimated optimal subset size, n=20). Age-based selection resulted in a similar marked improvement. We conclude that selecting atlases from large databases for atlas-based brain image segmentation improves the accuracy of the segmentations achieved. We show that image similarity is a suitable selection criterion and give results based on selecting atlases by age that demonstrate the value of meta-information for selection.

Amyloid, hypometabolism, and cognition in Alzheimer disease: An [11C]PIB and [18F]FDG PET study

Objective: To investigate the association between brain amyloid load in Alzheimer disease (AD) measured by [11C]PIB-PET, regional cerebral glucose metabolism (rCMRGlc) measured by [18F]FDG-PET, and cognition. Methods: Nineteen subjects with AD and 14 controls had [11C]PIB-PET and underwent a battery of psychometric tests. Twelve of those subjects with AD and eight controls had [18F]FDG-PET. Parametric images of [11C]PIB binding and rCMRGlc were interrogated with a region-of-interest atlas and statistical parametric mapping. [11C]PIB binding and rCMRGlc were correlated with scores on psychometric tests. Results: AD subjects showed twofold increases in mean [11C]PIB binding in cingulate, frontal, temporal, parietal, and occipital cortical areas. Higher cortical amyloid load correlated with lower scores on facial and word recognition tests. Two patients fulfilling the clinical criteria for AD had normal [11C]PIB at baseline. Over 20 months this remained normal in one but increased in the cingulate of the other. Mean levels of temporal and parietal rCMRGlc were reduced by 20% in AD and these correlated with mini mental scores, immediate recall, and recognition memory test for words. Higher [11C]PIB uptake correlated with lower rCMRGlc in temporal and parietal cortices. Conclusion: [11C]PIB-PET detected an increased amyloid plaque load in 89% of patients with clinically probable Alzheimer disease (AD). The high frontal amyloid load detected by [11C]PIB-PET in AD in the face of spared glucose metabolism is of interest and suggests that amyloid plaque formation may not be directly responsible for neuronal dysfunction in this disorder.

Diffeomorphic registration using b-splines

In this paper we propose a diffeomorphic non-rigid registration algorithm based on free-form deformations (FFDs) which are modelled by B-splines. In contrast to existing non-rigid registration methods based on FFDs the proposed diffeomorphic non-rigid registration algorithm based on free-form deformations (FFDs) which are modelled by B-splines. To construct a diffeomorphic transformation we compose a sequence of free-form deformations while ensuring that individual FFDs are one-to-one transformations. We have evaluated the algorithm on 20 normal brain MR images which have been manually segmented into 67 anatomical structures. Using the agreement between manual segmentation and segmentation propagation as a measure of registration quality we have compared the algorithm to an existing FFD registration algorithm and a modified FFD registration algorithm which penalises non-diffeomorphic transformations. The results show that the proposed algorithm generates diffeomorphic transformations while providing similar levels of performance as the existing FFD registration algorithm in terms of registration accuracy.

Automatic segmentation of brain MRIs of 2-year-olds into 83 regions of interest.

Three-dimensional atlases and databases of the brain at different ages facilitate the description of neuroanatomy and the monitoring of cerebral growth and development. Brain segmentation is challenging in young children due to structural differences compared to adults. We have developed a method, based on established algorithms, for automatic segmentation of young childrens brains into 83 regions of interest (ROIs), and applied this to an exemplar group of 33 2-year-old subjects who had been born prematurely. The algorithm uses prior information from 30 normal adult brain magnetic resonance (MR) images, which had been manually segmented to create 30 atlases, each labeling 83 anatomical structures. Each of these adult atlases was registered to each 2-year-old target MR image using non-rigid registration based on free-form deformations. Label propagation from each adult atlas yielded a segmentation of each 2-year-old brain into 83 ROIs. The final segmentation was obtained by combination of the 30 propagated adult atlases using decision fusion, improving accuracy over individual propagations. We validated this algorithm by comparing the automatic approach with three representative manually segmented volumetric regions (the subcortical caudate nucleus, the neocortical pre-central gyrus and the archicortical hippocampus) using similarity indices (SI), a measure of spatial overlap (intersection over average). SI results for automatic versus manual segmentations for these three structures were 0.90+/-0.01, 0.90+/-0.01 and 0.88+/-0.03 respectively. This registration approach allows the rapid construction of automatically labelled age-specific brain atlases for children at the age of 2 years.

Positron emission tomography Partial volume correction: estimation and algorithms

Partial volume effects in positron emission tomography (PET) lead to quantitative under- and over-estimations of the regional concentrations of radioactivity in reconstructed images and corresponding errors in derived functional or parametric images. The limited resolution of PET leads to “tissue-fraction” effects, reflecting underlying tissue heterogeneity, and “spillover” effects between regions. Addressing the former problem in general requires supplementary data, for example, coregistered high-resolution magnetic resonance images, whereas the latter effect can be corrected for with PET data alone if the point-spread function of the tomograph has been characterized. Analysis of otherwise homogeneous region-of-interest data ideally requires a combination of tissue classification and correction for the point-spread function. The formulation of appropriate algorithms for partial volume correction (PVC) is dependent on both the distribution of the signal and the distribution of the underlying noise. A mathematical framework has therefore been developed to accommodate both of these factors and to facilitate the development of new PVC algorithms based on the description of the problem. Several methodologies and algorithms have been proposed and implemented in the literature in order to address these problems. These methods do not, however, explicitly consider the noise model while differing in their underlying assumptions. The general theory for estimation of regional concentrations, associated error estimation, and inhomogeneity tests are presented in a weighted least squares framework. The analysis has been validated using both simulated and real PET data sets. The relations between the current algorithms and those published previously are formulated and compared. The incorporation of tensors into the formulation of the problem has led to the construction of computationally rapid algorithms taking into account both tissue-fraction and spillover effects. The suitability of their application to dynamic and static images is discussed.

Random forest-based similarity measures for multi-modal classification of Alzheimer's disease

Neurodegenerative disorders, such as Alzheimers disease, are associated with changes in multiple neuroimaging and biological measures. These may provide complementary information for diagnosis and prognosis. We present a multi-modality classification framework in which manifolds are constructed based on pairwise similarity measures derived from random forest classifiers. Similarities from multiple modalities are combined to generate an embedding that simultaneously encodes information about all the available features. Multi-modality classification is then performed using coordinates from this joint embedding. We evaluate the proposed framework by application to neuroimaging and biological data from the Alzheimers Disease Neuroimaging Initiative (ADNI). Features include regional MRI volumes, voxel-based FDG-PET signal intensities, CSF biomarker measures, and categorical genetic information. Classification based on the joint embedding constructed using information from all four modalities out-performs the classification based on any individual modality for comparisons between Alzheimers disease patients and healthy controls, as well as between mild cognitive impairment patients and healthy controls. Based on the joint embedding, we achieve classification accuracies of 89% between Alzheimers disease patients and healthy controls, and 75% between mild cognitive impairment patients and healthy controls. These results are comparable with those reported in other recent studies using multi-kernel learning. Random forests provide consistent pairwise similarity measures for multiple modalities, thus facilitating the combination of different types of feature data. We demonstrate this by application to data in which the number of features differs by several orders of magnitude between modalities. Random forest classifiers extend naturally to multi-class problems, and the framework described here could be applied to distinguish between multiple patient groups in the future.

Progressive neocortical damage in epilepsy

Our objective was to determine the pattern and extent of generalized and focal neocortical atrophy that develops in patients with epilepsy and the factors associated with such changes. As part of a prospective, longitudinal follow‐up study of 122 patients with chronic epilepsy, 68 newly diagnosed patients, and 90 controls, serial magnetic resonance imaging scans were obtained 3.5 years apart. Image subtraction was used to identify diffuse and focal neocortical change that was quantified with a regional brain atlas and a fully automated segmentation algorithm. New focal or generalized neocortical volume losses were identified in 54% of patients with chronic epilepsy, 39% of newly diagnosed patients and 24% of controls. Patients with chronic epilepsy were significantly more likely to develop neocortical atrophy than control subjects. The increased risk of cerebral atrophy in epilepsy was not related to a history of documented seizures. Risk factors for neocortical atrophy were age and multiple antiepileptic drug exposure. Focal and generalized neocortical atrophy commonly develops in chronic epilepsy. Neocortical changes seen in a quarter of our control group over 3.5 years were likely to reflect physiological changes. Our results show that ongoing cerebral atrophy may be widespread and remote from the putative epileptic focus, possibly reflecting extensive networks and interconnections between cortical regions. Ann Neurol 2003

Stroke following chiropractic manipulation of the cervical spine

Abstract We analyzed the clinical course and neuroradiological findings of ten patients aged 27–46 years, with ischemic stroke secondary to vertebral artery dissection (VAD; n = 8) or internal carotid artery dissection (CAD; n = 2), all following chiropractic manipulation of the cervical spine. The following observations were made: (a) All patients had uneventful medical histories, no or only mild vascular risk factors, and no predisposing vascular lesions. (b) VAD was unilateral in five patients and bilateral in three. VAD was located close to the atlantoaxial joint in all eight patients and showed additional involvement of lower sections in six, as well as temporary occlusion of one vertebral artery in three. (c) Nine of ten patients had brain infarction documented by magnetic resonance imaging or computed tomography. (d) Onset of symptoms was immediately after the manipulation (n = 5) or within 2 days (n = 5). (e) Progression of neurological deficits occurred within the following hours to a maximum of 3 weeks. (f) Maximum neurological deficits were severe in nine of ten patients. (g) Outcome after 4 weeks–3 years included no or mild neurological deficits in five patients, marked deficits in three, persistent locked-in syndrome in one, and persistent vegetative state in one. (h) Informed consent was obtained in only one of ten patients. Thus, patients at risk for stroke after chiropractic manipulation may not be identified a priori. Neurological deficits may be severely disabling and are potentially life threatening.